Assessing base-resolution DNA mechanics on the genome scale.

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• Intrinsic DNA properties, such as bending, are vital for biological functions.
• High-throughput loop-seq technology enables DNA bendability assessment but is challenging for large genomes.
• Base-resolution DNA bendability prediction is crucial for understanding genomic regulation.

Purpose of the Study:

• To introduce BendNet, a deep neural network for predicting base-resolution DNA bendability.
• To train BendNet using loop-seq data from yeast.
• To apply BendNet to large-scale genomic analyses and create a species-wide bendability resource.

Main Methods:

• Developed BendNet, a deep neural network model.
• Utilized loop-seq experimental data from yeast for model training.
• Applied BendNet to predict DNA bendability across the human genome and 307 other species.

Main Results:

• BendNet accurately predicts DNA bendability at base resolution.
• Predicted DNA bendability in the human genome correlates with chromatin features and disease-associated regions.
• Identified associations between DNA bendability and transcription regulation, DNA replication, transcription factor binding, and extrachromosomal circular DNA generation.

Conclusions:

• BendNet is a powerful tool for predicting intrinsic DNA bending.
• DNA bendability influences key genomic processes and disease risk in mammals.
• A comprehensive resource of genomic DNA bendability profiles for 307 species is now available, with an online tool for sequence analysis.